基于自适应遗传算法和SVM的特征选择

Feature Selection Based on Adaptive Genetic Algorithm and SVM

针对传统风险辨识方法无法实现盾构隧道施工过程中的风险状态实时识别的问题,提出一种自适应遗传算法和支持向量机结合的特征选择方法(AGASVM),筛选出与施工质量风险关系最为密切的关键特征集。实验结果表明,用AGASVM所获得的关键特征集用于施工风险状态实时识别的分类准确率较高。其特征集规模比原始特征集有明显缩减,而且绝大部分关键特征与领域专家的意见是吻合的。

Aiming at the question that the traditional method for discerning risk can not come true the real-time recognition of risk statue in the shield tunneling constructing process, this paper proposes a feature selection method which combines Adaptive Genetic Algorithm with Support Vector Machine（AGASVM）. It is used to filter a pivotal feature subset which is super correlative with risk of constructing quality. Experimental result shows that the pivotal feature subset selected by AGASVM can make the classification accuracy higher when it is used in the real-time recognition of risk statue. The dimension of pivotal feature subset is obviously smaller than the one of original factors set, and the most of pivotal features are the same as the ideas of domain experts.